IPTV App

Most IPTV apps optimize for channel count, not for actually finding something to watch, clunky grids, no personalization, and metadata that's often wrong or missing entirely.

Build an Android TV app where AI handles the tedious parts, generating clean metadata and organizing content around what someone's actually likely to want next, while keeping the interface fast and simple.

A thin Android TV client talks to a backend that ingests stream sources, normalizes metadata through an LLM pass, and serves a personalized catalog via a lightweight API, no heavyweight media server in the critical path.

A working Android TV app with live channel browsing, AI-generated show/movie metadata (titles, descriptions, artwork matching), and a recommendation feed based on viewing patterns.

Coding agents handled most of the Android/Kotlin UI scaffolding and API integration; a separate LLM pipeline generates and cleans metadata for content that arrives with inconsistent or missing labels.

Kept metadata generation as an offline batch job rather than real-time, to keep the app snappy and avoid paying LLM latency on every screen load.

The metadata pipeline dramatically improved browse quality, mislabeled or blank entries basically disappeared.

Early attempts at real-time AI recommendations were too slow and too expensive to run per request; moved to precomputed batches instead.

For a TV app, perceived speed matters more than freshness, batch-and-cache beats real-time almost everywhere in the UI.